What’s new
- Tuned for information-dense images: Agnes AI says this version handles complex compositions and detail-rich scenes — posters, infographic-style images — better than 2.0.
- Resolution moved to tiers plus ratio: instead of typing exact pixel dimensions, you pick a 1K/2K/3K/4K tier and one of 8 aspect ratios (1:1, 3:4, 4:3, 16:9, 9:16, 2:3, 3:2, 21:9); 2.0 still takes literal pixel sizes.
- Better composition preservation on edits: image-to-image and local edits hold the original layout together more reliably instead of drifting.
- Multi-image composition: beyond editing one image, it can combine several reference images into one new output.
What it’s good for
- Information-dense visuals: posters, infographics, layout-heavy marketing assets.
- Edits that need to keep the original composition — restyling without reflowing the layout.
- Multi-image composition: stitching several assets into a single new image.
Prompting tips
- Structure the prompt: Agnes AI's own recommended order is subject + scene/environment + style + lighting + composition + quality requirements — going through them one at a time beats stacking adjectives.
- For edits, say what changes and what stays: in image-to-image or editing tasks, spell out both sides explicitly, or the model will decide on its own.
- Reference images must be publicly reachable: if you pass a URL, it has to load without authentication — no gated links.
- Give each image a role in multi-image composition: name what each reference contributes to the final shot to avoid the model mixing up which is which.
- A bigger tier isn't automatically the right choice: 4K suits print-grade assets; 1K/2K is usually enough for everyday images — pick by use case rather than maxing it out by default.
Subject: a cup of hand-poured coffee with steam rising, on a wooden table. Scene: by a window at dawn, soft natural light from the left.
Style: still-life photography, shallow depth of field. Composition: subject centered-left, negative space on the right for text. Quality: sharp, detail-rich, 4K.
Known limits
- An exact pixel size outside the tier list (e.g. 1920x1080) gets normalized to the nearest tier rather than produced as-is.
- Reference image URLs must be publicly accessible; authenticated links aren't supported.
- Processing time scales with scene complexity; Agnes AI recommends clients allow a 60-360 second timeout.
Using it in Nomi
On the generation canvas, add a image card, pick “Agnes Image 2.1” in the model picker and choose a mode (Text to image, Image editing). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
